static void process_gpu_request_reply(
struct work_task *pwt)
{
char *id = "process_gpu_request_reply";
struct batch_request *preq;
svr_disconnect(pwt->wt_event); /* close connection to MOM */
preq = pwt->wt_parm1;
preq->rq_conn = preq->rq_orgconn; /* restore client socket */
if (preq->rq_reply.brp_code != 0)
{
sprintf(log_buffer,
"MOM failed on GPU request, rc = %d",
preq->rq_reply.brp_code);
log_err(errno, id, log_buffer);
req_reject(preq->rq_reply.brp_code, 0, preq, NULL, log_buffer);
}
else
{
/* record that MOM changed gpu mode */
if (LOGLEVEL >= 7)
{
sprintf(
log_buffer,
"GPU control request completed for node %s gpuid %s mode %d reset_perm %d reset_vol %d",
preq->rq_ind.rq_gpuctrl.rq_momnode,
preq->rq_ind.rq_gpuctrl.rq_gpuid,
preq->rq_ind.rq_gpuctrl.rq_gpumode,
preq->rq_ind.rq_gpuctrl.rq_reset_perm,
preq->rq_ind.rq_gpuctrl.rq_reset_vol);
log_ext(-1, id, log_buffer, LOG_INFO);
}
reply_ack(preq);
}
}
void req_jobcredential(
struct batch_request *preq) /* ptr to the decoded request */
{
job *pj;
pj = locate_new_job(preq->rq_conn, NULL);
if (pj == NULL)
{
req_reject(PBSE_IVALREQ, 0, preq, NULL, NULL);
return;
}
reply_ack(preq);
return;
} /* END req_jobcredential() */
/*
* process_gpu_request_reply
* called when a gpu change request was sent to MOM and the answer
* is received. Completes the gpu request.
*/
void process_gpu_request_reply(
batch_request *preq)
{
char log_buf[LOCAL_LOG_BUF_SIZE];
if (preq == NULL)
return;
preq->rq_conn = preq->rq_orgconn; /* restore client socket */
if (preq->rq_reply.brp_code != 0)
{
sprintf(log_buf,
"MOM failed on GPU request, rc = %d",
preq->rq_reply.brp_code);
log_err(errno, __func__, log_buf);
req_reject(preq->rq_reply.brp_code, 0, preq, NULL, log_buf);
}
else
{
/* record that MOM changed gpu mode */
if (LOGLEVEL >= 7)
{
sprintf(
log_buf,
"GPU control request completed for node %s gpuid %s mode %d reset_perm %d reset_vol %d",
preq->rq_ind.rq_gpuctrl.rq_momnode,
preq->rq_ind.rq_gpuctrl.rq_gpuid,
preq->rq_ind.rq_gpuctrl.rq_gpumode,
preq->rq_ind.rq_gpuctrl.rq_reset_perm,
preq->rq_ind.rq_gpuctrl.rq_reset_vol);
log_ext(-1, __func__, log_buf, LOG_INFO);
}
reply_ack(preq);
}
} /* END process_gpu_request_reply() */
int get_UID(
int s,
char *munge_buf,
struct batch_request *preq)
{
char *ptr;
char user_name[PBS_MAXUSER];
int i = 0;
ptr = strstr(munge_buf, "UID:");
if (!ptr)
{
req_reject(PBSE_SYSTEM, 0, preq, NULL, "could not read unmunge data user");
return(-1);
}
ptr = strchr(ptr, ':');
ptr++;
while (*ptr == SPACE)
{
ptr++;
}
memset(user_name, 0, sizeof(user_name));
while ((*ptr != SPACE) &&
(!isspace(*ptr)) &&
(i < (int)sizeof(user_name)))
{
user_name[i++] = *ptr;
ptr++;
}
strncpy(conn_credent[s].username, user_name, sizeof(conn_credent[s].username) - 1);
conn_credent[s].username[sizeof(conn_credent[s].username) - 1] = 0;
return(PBSE_NONE);
} /* END get_UID() */
void post_message_req(
batch_request *preq)
{
char log_buf[LOCAL_LOG_BUF_SIZE];
/* preq has been hadnled previously */
if (preq == NULL)
return;
preq->rq_conn = preq->rq_orgconn; /* restore socket to client */
sprintf(log_buf, msg_messagejob, preq->rq_reply.brp_code);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, preq->rq_ind.rq_message.rq_jid, log_buf);
if (preq->rq_reply.brp_code)
req_reject(preq->rq_reply.brp_code, 0, preq, NULL, NULL);
else
reply_ack(preq);
} /* END post_message_req() */
void *modify_job_work(
batch_request *vp) /* I */
{
job *pjob;
svrattrl *plist;
int checkpoint_req = FALSE;
batch_request *preq = (struct batch_request *)vp;
pjob = svr_find_job(preq->rq_ind.rq_modify.rq_objname, FALSE);
if (pjob == NULL)
{
req_reject(PBSE_JOBNOTFOUND, 0, preq, NULL, "Job unexpectedly deleted");
return(NULL);
}
mutex_mgr job_mutex(pjob->ji_mutex, true);
/* pbs_mom sets the extend string to trigger copying of checkpoint files */
if (preq->rq_extend != NULL)
{
if (strcmp(preq->rq_extend,CHECKPOINTHOLD) == 0)
{
checkpoint_req = CHK_HOLD;
}
else if (strcmp(preq->rq_extend,CHECKPOINTCONT) == 0)
{
checkpoint_req = CHK_CONT;
}
}
plist = (svrattrl *)GET_NEXT(preq->rq_ind.rq_modify.rq_attr);
/* modify_job will free preq and respond to it */
modify_job((void **)&pjob, plist, preq, checkpoint_req, 0);
return(NULL);
} /* END modify_job_work() */
static void process_checkpoint_reply(
struct work_task *pwt)
{
job *pjob;
struct batch_request *preq;
svr_disconnect(pwt->wt_event); /* close connection to MOM */
preq = get_remove_batch_request(pwt->wt_parm1);
free(pwt->wt_mutex);
free(pwt);
/* preq handled previously */
if (preq == NULL)
return;
preq->rq_conn = preq->rq_orgconn; /* restore client socket */
if ((pjob = svr_find_job(preq->rq_ind.rq_manager.rq_objname, FALSE)) == NULL)
{
log_event(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB,
preq->rq_ind.rq_manager.rq_objname,
msg_postmomnojob);
req_reject(PBSE_UNKJOBID, 0, preq, NULL, msg_postmomnojob);
}
else
{
/* record that MOM has a checkpoint file */
account_record(PBS_ACCT_CHKPNT, pjob, "Checkpointed"); /* note in accounting file */
reply_ack(preq);
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
}
} /* END process_checkpoint_reply() */
void
req_shutdown(struct batch_request *preq)
{
int type;
extern int shutdown_who;
if ((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_MGRD | ATR_DFLAG_OPRD |
ATR_DFLAG_OPWR)) == 0) {
req_reject(PBSE_PERM, 0, preq);
return;
}
(void)sprintf(log_buffer, msg_shutdown_op, preq->rq_user, preq->rq_host);
log_event(PBSEVENT_SYSTEM|PBSEVENT_ADMIN|PBSEVENT_DEBUG,
PBS_EVENTCLASS_SERVER, LOG_NOTICE, msg_daemonname, log_buffer);
pshutdown_request = preq; /* save for reply from main() when done */
type = preq->rq_ind.rq_shutdown;
shutdown_who = type & SHUT_WHO_MASK;
if (shutdown_who & SHUT_WHO_SECDONLY)
(void)failover_send_shutdown(FAILOVER_SecdShutdown);
if (shutdown_who & SHUT_WHO_SCHED)
(void)contact_sched(SCH_QUIT, NULL); /* tell scheduler to quit */
if (shutdown_who & SHUT_WHO_SECDONLY) {
reply_ack(preq);
return; /* do NOT shutdown this Server */
}
/* Moms are told to shutdown in pbsd_main.c after main loop */
svr_shutdown(type);
return;
}
void
req_stat_sched(struct batch_request *preq)
{
svrattrl *pal;
struct batch_reply *preply;
int rc = 0;
pbs_sched *psched;
/* allocate a reply structure and a status sub-structure */
preply = &preq->rq_reply;
preply->brp_choice = BATCH_REPLY_CHOICE_Status;
CLEAR_HEAD(preply->brp_un.brp_status);
for (psched = (pbs_sched *) GET_NEXT(svr_allscheds);
(psched != NULL);
psched = (pbs_sched *) GET_NEXT(psched->sc_link)
) {
rc = status_sched(psched, preq, &preply->brp_un.brp_status);
if (rc != 0) {
break;
}
}
if (!rc) {
(void)reply_send(preq);
} else {
if (rc != PBSE_NOATTR)
req_reject(rc, 0, preq);
else {
pal = (svrattrl *)GET_NEXT(preq->rq_ind.
rq_status.rq_attr);
reply_badattr(rc, bad, pal, preq);
}
}
}
int handle_single_delete(
batch_request *preq,
batch_request *preq_tmp,
char *Msg)
{
char *jobid = preq->rq_ind.rq_delete.rq_objname;
job *pjob = svr_find_job(jobid, FALSE);
if (pjob == NULL)
{
log_event(PBSEVENT_DEBUG,PBS_EVENTCLASS_JOB,jobid,pbse_to_txt(PBSE_UNKJOBID));
req_reject(PBSE_UNKJOBID, 0, preq, NULL, "cannot locate job");
}
else
{
std::string jobID = pjob->ji_qs.ji_jobid;
unlock_ji_mutex(pjob, __func__, NULL, LOGLEVEL);
removeBeforeAnyDependencies(jobID.c_str());
/* send the asynchronous reply if needed */
if (preq_tmp != NULL)
{
reply_ack(preq_tmp);
preq->rq_noreply = TRUE; /* set for no more replies */
enqueue_threadpool_request(single_delete_work, preq, async_pool);
}
else
single_delete_work(preq);
}
return(PBSE_NONE);
} /* END handle_single_delete() */
void req_connect(
struct batch_request *preq)
{
int sock = preq->rq_conn;
unsigned short conn_authen;
/* Called from one location inside a lock */
pthread_mutex_lock(svr_conn[sock].cn_mutex);
conn_authen = svr_conn[sock].cn_authen;
pthread_mutex_unlock(svr_conn[sock].cn_mutex);
if (conn_authen == 0)
{
reply_ack(preq);
}
else
{
req_reject(PBSE_BADCRED, 0, preq, NULL, "Connection not authorized");
}
return;
} /* END req_connect() */
int handle_single_delete(
struct batch_request *preq,
struct batch_request *preq_tmp,
char *Msg)
{
int rc= -1;
char *jobid = preq->rq_ind.rq_delete.rq_objname;
job *pjob = svr_find_job(jobid, FALSE);
if (pjob == NULL)
{
log_event(PBSEVENT_DEBUG,PBS_EVENTCLASS_JOB,jobid,pbse_to_txt(PBSE_UNKJOBID));
req_reject(PBSE_UNKJOBID, 0, preq, NULL, "cannot locate job");
}
else
{
if (preq_tmp != NULL)
{
reply_ack(preq_tmp);
preq->rq_noreply = TRUE; /* set for no more replies */
}
/* mutex is freed below */
if ((rc = forced_jobpurge(pjob, preq)) == PBSE_NONE)
rc = execute_job_delete(pjob, Msg, preq);
}
if ((rc == PBSE_NONE) ||
(rc == PURGE_SUCCESS))
reply_ack(preq);
return(PBSE_NONE);
} /* END handle_single_delete() */
static void req_stat_job_step2(
struct stat_cntl *cntl) /* I/O (free'd on return) */
{
svrattrl *pal;
job *pjob = NULL;
struct batch_request *preq;
struct batch_reply *preply;
int rc = 0;
enum TJobStatTypeEnum type;
pbs_queue *pque = NULL;
int exec_only = 0;
int bad = 0;
long DTime; /* delta time - only report full pbs_attribute list if J->MTime > DTime */
static svrattrl *dpal = NULL;
int job_array_index = 0;
job_array *pa = NULL;
char log_buf[LOCAL_LOG_BUF_SIZE];
all_jobs_iterator *iter;
preq = cntl->sc_origrq;
type = (enum TJobStatTypeEnum)cntl->sc_type;
preply = &preq->rq_reply;
/* See pbs_server_attributes(1B) for details on "poll_jobs" behaviour */
if (dpal == NULL)
{
/* build 'delta' pbs_attribute list */
svrattrl *tpal;
tlist_head dalist;
int aindex;
int atrlist[] =
{
JOB_ATR_jobname,
JOB_ATR_resc_used,
JOB_ATR_LAST
};
CLEAR_LINK(dalist);
for (aindex = 0;atrlist[aindex] != JOB_ATR_LAST;aindex++)
{
if ((tpal = attrlist_create("", "", 23)) == NULL)
{
return;
}
tpal->al_valln = atrlist[aindex];
if (dpal == NULL)
dpal = tpal;
append_link(&dalist, &tpal->al_link, tpal);
}
} /* END if (dpal == NULL) */
if (type == tjstArray)
{
pa = get_array(preq->rq_ind.rq_status.rq_id);
if (pa == NULL)
{
req_reject(PBSE_UNKARRAYID, 0, preq, NULL, "unable to find array");
return;
}
}
{
all_jobs *ajptr = NULL;
if (type == tjstQueue)
ajptr = cntl->sc_pque->qu_jobs;
else if (type == tjstSummarizeArraysQueue)
ajptr = cntl->sc_pque->qu_jobs_array_sum;
else if (type == tjstSummarizeArraysServer)
ajptr = &array_summary;
else
ajptr = &alljobs;
ajptr->lock();
iter = ajptr->get_iterator();
ajptr->unlock();
}
/*
* now ready for part 3, building the status reply,
* loop through again
*/
if ((type == tjstSummarizeArraysQueue) ||
(type == tjstSummarizeArraysServer))
{
/* No array can be owned for these options */
update_array_statuses();
}
if (type == tjstJob)
pjob = svr_find_job(preq->rq_ind.rq_status.rq_id, FALSE);
else if (type == tjstQueue)
pjob = next_job(cntl->sc_pque->qu_jobs,iter);
else if (type == tjstSummarizeArraysQueue)
pjob = next_job(cntl->sc_pque->qu_jobs_array_sum,iter);
else if (type == tjstSummarizeArraysServer)
pjob = next_job(&array_summary,iter);
else if (type == tjstArray)
{
job_array_index = -1;
pjob = NULL;
/* increment job_array_index until we find a non-null pointer or hit the end */
while (++job_array_index < pa->ai_qs.array_size)
{
if (pa->job_ids[job_array_index] != NULL)
{
if ((pjob = svr_find_job(pa->job_ids[job_array_index], FALSE)) != NULL)
{
break;
}
}
}
}
else
pjob = next_job(&alljobs,iter);
DTime = 0;
if (preq->rq_extend != NULL)
{
char *ptr;
/* FORMAT: { EXECQONLY | DELTA:<EPOCHTIME> } */
if (strstr(preq->rq_extend, EXECQUEONLY))
exec_only = 1;
ptr = strstr(preq->rq_extend, "DELTA:");
if (ptr != NULL)
{
ptr += strlen("delta:");
DTime = strtol(ptr, NULL, 10);
}
}
if ((type == tjstTruncatedServer) ||
(type == tjstTruncatedQueue))
{
long sentJobCounter;
long qjcounter;
long qmaxreport;
all_queues_iterator *iter = NULL;
svr_queues.lock();
iter = svr_queues.get_iterator();
svr_queues.unlock();
/* loop through all queues */
while ((pque = next_queue(&svr_queues,iter)) != NULL)
{
qjcounter = 0;
if ((exec_only == 1) &&
(pque->qu_qs.qu_type != QTYPE_Execution))
{
/* ignore routing queues */
unlock_queue(pque, __func__, "ignore queue", LOGLEVEL);
continue;
}
if (((pque->qu_attr[QA_ATR_MaxReport].at_flags & ATR_VFLAG_SET) != 0) &&
(pque->qu_attr[QA_ATR_MaxReport].at_val.at_long >= 0))
{
qmaxreport = pque->qu_attr[QA_ATR_MaxReport].at_val.at_long;
}
else
{
qmaxreport = TMAX_JOB;
}
if (LOGLEVEL >= 5)
{
sprintf(log_buf,"giving scheduler up to %ld idle jobs in queue %s\n",
qmaxreport,
pque->qu_qs.qu_name);
log_event(PBSEVENT_SYSTEM,PBS_EVENTCLASS_QUEUE,pque->qu_qs.qu_name,log_buf);
}
sentJobCounter = 0;
/* loop through jobs in queue */
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, "5", LOGLEVEL);
all_jobs_iterator *jobiter = NULL;
pque->qu_jobs->lock();
jobiter = pque->qu_jobs->get_iterator();
pque->qu_jobs->unlock();
while ((pjob = next_job(pque->qu_jobs,jobiter)) != NULL)
{
if ((qjcounter >= qmaxreport) &&
(pjob->ji_qs.ji_state == JOB_STATE_QUEUED))
{
/* max_report of queued jobs reached for queue */
unlock_ji_mutex(pjob, __func__, "6", LOGLEVEL);
continue;
}
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
rc = status_job(
pjob,
preq,
(pjob->ji_wattr[JOB_ATR_mtime].at_val.at_long >= DTime) ? pal : dpal,
&preply->brp_un.brp_status,
&bad);
if ((rc != 0) && (rc != PBSE_PERM))
{
req_reject(rc, bad, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, "7", LOGLEVEL);
unlock_queue(pque, __func__, "perm", LOGLEVEL);
delete iter;
return;
}
sentJobCounter++;
if (pjob->ji_qs.ji_state == JOB_STATE_QUEUED)
qjcounter++;
unlock_ji_mutex(pjob, __func__, "8", LOGLEVEL);
} /* END foreach (pjob from pque) */
if (LOGLEVEL >= 5)
{
sprintf(log_buf,"sent scheduler %ld total jobs for queue %s\n",
sentJobCounter,
pque->qu_qs.qu_name);
log_event(PBSEVENT_SYSTEM,PBS_EVENTCLASS_QUEUE,pque->qu_qs.qu_name,log_buf);
}
unlock_queue(pque, __func__, "end while", LOGLEVEL);
} /* END for (pque) */
reply_send_svr(preq);
delete iter;
return;
} /* END if ((type == tjstTruncatedServer) || ...) */
while (pjob != NULL)
{
/* go ahead and build the status reply for this job */
if (exec_only)
{
if (cntl->sc_pque != NULL)
{
if (cntl->sc_pque->qu_qs.qu_type != QTYPE_Execution)
goto nextjob;
}
else
{
if (pa != NULL)
pthread_mutex_unlock(pa->ai_mutex);
pque = get_jobs_queue(&pjob);
if (pa != NULL)
pthread_mutex_lock(pa->ai_mutex);
if ((pjob == NULL) ||
(pque == NULL))
goto nextjob;
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
if (pque->qu_qs.qu_type != QTYPE_Execution)
{
goto nextjob;
}
}
}
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
rc = status_job(
pjob,
preq,
pal,
&preply->brp_un.brp_status,
&bad);
if ((rc != 0) &&
(rc != PBSE_PERM))
{
if (pa != NULL)
{
unlock_ai_mutex(pa, __func__, "1", LOGLEVEL);
}
unlock_ji_mutex(pjob, __func__, "9", LOGLEVEL);
req_reject(rc, bad, preq, NULL, NULL);
delete iter;
return;
}
/* get next job */
nextjob:
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, "10", LOGLEVEL);
if (type == tjstJob)
break;
if (type == tjstQueue)
pjob = next_job(cntl->sc_pque->qu_jobs,iter);
else if (type == tjstSummarizeArraysQueue)
pjob = next_job(cntl->sc_pque->qu_jobs_array_sum,iter);
else if (type == tjstSummarizeArraysServer)
pjob = next_job(&array_summary,iter);
else if (type == tjstArray)
{
pjob = NULL;
/* increment job_array_index until we find a non-null pointer or hit the end */
while (++job_array_index < pa->ai_qs.array_size)
{
if (pa->job_ids[job_array_index] != NULL)
{
if ((pjob = svr_find_job(pa->job_ids[job_array_index], FALSE)) != NULL)
{
break;
}
}
}
}
else
pjob = next_job(&alljobs,iter);
rc = 0;
} /* END while (pjob != NULL) */
delete iter;
if (pa != NULL)
{
unlock_ai_mutex(pa, __func__, "1", LOGLEVEL);
}
reply_send_svr(preq);
if (LOGLEVEL >= 7)
{
log_event(PBSEVENT_SYSTEM,
PBS_EVENTCLASS_JOB,
"req_statjob",
"Successfully returned the status of queued jobs\n");
}
return;
} /* END req_stat_job_step2() */
int execute_job_delete(
job *pjob, /* M */
char *Msg, /* I */
struct batch_request *preq) /* I */
{
struct work_task *pwtnew;
int rc;
char *sigt = "SIGTERM";
int has_mutex = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
time_t time_now = time(NULL);
long force_cancel = FALSE;
long array_compatible = FALSE;
chk_job_req_permissions(&pjob,preq);
if (pjob == NULL)
{
/* preq is rejected in chk_job_req_permissions here */
return(-1);
}
if (pjob->ji_qs.ji_state == JOB_STATE_TRANSIT)
{
/* see note in req_delete - not sure this is possible still,
* but the deleted code is irrelevant now. I will leave this
* part --dbeer */
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return(-1);
}
if (pjob->ji_qs.ji_substate == JOB_SUBSTATE_PRERUN ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN1 ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN2 ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN3 )
{
/* If JOB_SUBSTATE_PRERUN being sent to MOM, wait till she gets it going */
/* retry in one second */
/* If JOB_SUBSTATE_RERUN, RERUN1, RERUN2 or RERUN3 the
job is being requeued. Wait until finished */
static time_t cycle_check_when = 0;
static char cycle_check_jid[PBS_MAXSVRJOBID + 1];
if (cycle_check_when != 0)
{
if (!strcmp(pjob->ji_qs.ji_jobid, cycle_check_jid) &&
(time_now - cycle_check_when > 10))
{
/* state not updated after 10 seconds */
/* did the mom ever get it? delete it anyways... */
cycle_check_jid[0] = '\0';
cycle_check_when = 0;
goto jump;
}
if (time_now - cycle_check_when > 20)
{
/* give up after 20 seconds */
cycle_check_jid[0] = '\0';
cycle_check_when = 0;
}
} /* END if (cycle_check_when != 0) */
if (cycle_check_when == 0)
{
/* new PRERUN job located */
cycle_check_when = time_now;
strcpy(cycle_check_jid, pjob->ji_qs.ji_jobid);
}
sprintf(log_buf, "job cannot be deleted, state=PRERUN, requeuing delete request");
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
pwtnew = set_task(WORK_Timed,time_now + 1,post_delete_route,preq,FALSE);
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
if (pwtnew == NULL)
{
req_reject(PBSE_SYSTEM, 0, preq, NULL, NULL);
return(-1);
}
else
{
return(ROUTE_DELETE);
}
} /* END if (pjob->ji_qs.ji_substate == JOB_SUBSTATE_PRERUN) */
jump:
/*
* Log delete and if requesting client is not job owner, send mail.
*/
sprintf(log_buf, "requestor=%s@%s", preq->rq_user, preq->rq_host);
/* NOTE: should annotate accounting record with extend message (NYI) */
account_record(PBS_ACCT_DEL, pjob, log_buf);
sprintf(log_buf, msg_manager, msg_deletejob, preq->rq_user, preq->rq_host);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
/* NOTE: should incorporate job delete message */
if (Msg != NULL)
{
/* have text message in request extension, add it */
strcat(log_buf, "\n");
strcat(log_buf, Msg);
}
if ((svr_chk_owner(preq, pjob) != 0) &&
(pjob->ji_has_delete_nanny == FALSE))
{
/* only send email if owner did not delete job and job deleted
has not been previously attempted */
svr_mailowner(pjob, MAIL_DEL, MAIL_FORCE, log_buf);
/*
* If we sent mail and already sent the extra message
* then reset message so we don't trigger a redundant email
* in job_abt()
*/
if (Msg != NULL)
{
Msg = NULL;
}
}
if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) != 0)
{
/* job has restart file at mom, change restart comment if failed */
change_restart_comment_if_needed(pjob);
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/*
* setup a nanny task to make sure the job is actually deleted (see the
* comments at job_delete_nanny()).
*/
if (pjob->ji_has_delete_nanny == TRUE)
{
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
req_reject(PBSE_IVALREQ, 0, preq, NULL, "job cancel in progress");
return(-1);
}
apply_job_delete_nanny(pjob, time_now + 60);
/*
* Send signal request to MOM. The server will automagically
* pick up and "finish" off the client request when MOM replies.
*/
get_batch_request_id(preq);
if ((rc = issue_signal(&pjob, sigt, post_delete_mom1, strdup(preq->rq_id))))
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
/* normally will ack reply when mom responds */
if (pjob != NULL)
{
sprintf(log_buf, msg_delrunjobsig, sigt);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
}
return(-1);
} /* END if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING) */
/* make a cleanup task if set */
get_svr_attr_l(SRV_ATR_JobForceCancelTime, &force_cancel);
if (force_cancel > 0)
{
char *dup_jobid = strdup(pjob->ji_qs.ji_jobid);
set_task(WORK_Timed, time_now + force_cancel, ensure_deleted, dup_jobid, FALSE);
}
/* if configured, and this job didn't have a slot limit hold, free a job
* held with the slot limit hold */
get_svr_attr_l(SRV_ATR_MoabArrayCompatible, &array_compatible);
if ((array_compatible != FALSE) &&
((pjob->ji_wattr[JOB_ATR_hold].at_val.at_long & HOLD_l) == FALSE))
{
if ((pjob->ji_arraystruct != NULL) &&
(pjob->ji_is_array_template == FALSE))
{
int i;
int newstate;
int newsub;
job *tmp;
job_array *pa = get_jobs_array(&pjob);
if (pjob == NULL)
return(-1);
for (i = 0; i < pa->ai_qs.array_size; i++)
{
if (pa->job_ids[i] == NULL)
continue;
if (!strcmp(pa->job_ids[i], pjob->ji_qs.ji_jobid))
continue;
if ((tmp = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
if (tmp->ji_wattr[JOB_ATR_hold].at_val.at_long & HOLD_l)
{
tmp->ji_wattr[JOB_ATR_hold].at_val.at_long &= ~HOLD_l;
if (tmp->ji_wattr[JOB_ATR_hold].at_val.at_long == 0)
{
tmp->ji_wattr[JOB_ATR_hold].at_flags &= ~ATR_VFLAG_SET;
}
svr_evaljobstate(tmp, &newstate, &newsub, 1);
svr_setjobstate(tmp, newstate, newsub, FALSE);
job_save(tmp, SAVEJOB_FULL, 0);
unlock_ji_mutex(tmp, __func__, "5", LOGLEVEL);
break;
}
unlock_ji_mutex(tmp, __func__, "6", LOGLEVEL);
}
}
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "%s: unlocking ai_mutex", __func__);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
}
pthread_mutex_unlock(pa->ai_mutex);
}
} /* END MoabArrayCompatible check */
if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) != 0)
{
/* job has restart file at mom, do end job processing */
svr_setjobstate(pjob, JOB_STATE_EXITING, JOB_SUBSTATE_EXITING, FALSE);
/* force new connection */
pjob->ji_momhandle = -1;
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "calling on_job_exit from %s", __func__);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
}
set_task(WORK_Immed, 0, on_job_exit, strdup(pjob->ji_qs.ji_jobid), FALSE);
}
else if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn) != 0)
{
/* job has staged-in file, should remove them */
remove_stagein(&pjob);
if (pjob != NULL)
job_abt(&pjob, Msg);
has_mutex = FALSE;
}
else
{
/*
* the job is not transitting (though it may have been) and
* is not running, so put in into a complete state.
*/
struct pbs_queue *pque;
int KeepSeconds = 0;
svr_setjobstate(pjob, JOB_STATE_COMPLETE, JOB_SUBSTATE_COMPLETE, FALSE);
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
pque->qu_numcompleted++;
unlock_queue(pque, __func__, NULL, LOGLEVEL);
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "calling on_job_exit from %s", __func__);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
}
pthread_mutex_lock(server.sv_attr_mutex);
KeepSeconds = attr_ifelse_long(
&pque->qu_attr[QE_ATR_KeepCompleted],
&server.sv_attr[SRV_ATR_KeepCompleted],
0);
pthread_mutex_unlock(server.sv_attr_mutex);
}
else
KeepSeconds = 0;
if (pjob != NULL)
{
set_task(WORK_Timed, time_now + KeepSeconds, on_job_exit, strdup(pjob->ji_qs.ji_jobid), FALSE);
}
else
has_mutex = FALSE;
} /* END else if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) != 0) */
if (has_mutex == TRUE)
unlock_ji_mutex(pjob, __func__, "7", LOGLEVEL);
return(PBSE_NONE);
} /* END execute_job_delete() */
int handle_delete_all(
struct batch_request *preq,
struct batch_request *preq_tmp,
char *Msg)
{
/* don't use the actual request so we can reply about all of the jobs */
struct batch_request *preq_dup = duplicate_request(preq);
job *pjob;
int iter = -1;
int failed_deletes = 0;
int total_jobs = 0;
int rc = PBSE_NONE;
char tmpLine[MAXLINE];
preq_dup->rq_noreply = TRUE;
if (preq_tmp != NULL)
{
reply_ack(preq_tmp);
preq->rq_noreply = TRUE; /* set for no more replies */
}
while ((pjob = next_job(&alljobs, &iter)) != NULL)
{
if ((rc = forced_jobpurge(pjob, preq_dup)) == PURGE_SUCCESS)
{
continue;
}
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
continue;
}
total_jobs++;
/* mutex is freed below */
if (rc == PBSE_NONE)
{
if ((rc = execute_job_delete(pjob, Msg, preq_dup)) == PBSE_NONE)
reply_ack(preq_dup);
/* mark this as NULL because it has been freed */
preq_dup = NULL;
}
if (rc != PURGE_SUCCESS)
{
/* duplicate the preq so we don't have a problem with double frees */
preq_dup = duplicate_request(preq);
preq_dup->rq_noreply = TRUE;
if ((rc == MOM_DELETE) ||
(rc == ROUTE_DELETE))
failed_deletes++;
}
}
if (failed_deletes == 0)
{
reply_ack(preq);
/* PURGE SUCCESS means this was qdel -p all. In this case no reply_*()
* functions have been called */
if (rc == PURGE_SUCCESS)
{
free_br(preq_dup);
preq_dup = NULL;
}
}
else
{
snprintf(tmpLine,sizeof(tmpLine),"Deletes failed for %d of %d jobs",
failed_deletes,
total_jobs);
req_reject(PBSE_SYSTEM, 0, preq, NULL, tmpLine);
}
/* preq_dup happens at the end of the loop, so free the extra one if
* it is there */
if (preq_dup != NULL)
free_br(preq_dup);
return(PBSE_NONE);
} /* END handle_delete_all() */
int finalize_rerunjob(
batch_request *preq,
job *pjob,
int rc)
{
int Force;
char log_buf[LOCAL_LOG_BUF_SIZE];
if (pjob == NULL)
return(PBSE_BAD_PARAMETER);
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
if (preq->rq_extend && !strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE)))
Force = 1;
else
Force = 0;
switch (rc)
{
case -1:
/* completed job was requeued */
/* clear out job completion time if there is one */
break;
case 0:
/* requeue request successful */
pjob->ji_qs.ji_substate = JOB_SUBSTATE_RERUN;
break;
case PBSE_SYSTEM: /* This may not be accurate...*/
rc = PBSE_MEM_MALLOC;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "Can not allocate memory");
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
break;
default:
if (Force == 0)
{
rc = PBSE_MOMREJECT;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "Rejected by mom");
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
else
{
int newstate;
int newsubst;
unsigned int dummy;
char *tmp;
if ((cray_enabled == true) &&
(pjob->ji_wattr[JOB_ATR_login_node_id].at_val.at_str != NULL))
tmp = parse_servername(pjob->ji_wattr[JOB_ATR_login_node_id].at_val.at_str, &dummy);
else
tmp = parse_servername(pjob->ji_wattr[JOB_ATR_exec_host].at_val.at_str, &dummy);
/* Cannot communicate with MOM, forcibly requeue job.
This is a relatively disgusting thing to do */
sprintf(log_buf, "rerun req to %s failed (rc=%d), forcibly requeueing job",
tmp, rc);
free(tmp);
log_event(
PBSEVENT_ERROR | PBSEVENT_ADMIN | PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buf);
log_err(-1, __func__, log_buf);
strcat(log_buf, ", previous output files may be lost");
svr_mailowner(pjob, MAIL_OTHER, MAIL_FORCE, log_buf);
svr_setjobstate(pjob, JOB_STATE_EXITING, JOB_SUBSTATE_RERUN3, FALSE);
rel_resc(pjob); /* free resc assigned to job */
pjob->ji_modified = 1; /* force full job save */
pjob->ji_momhandle = -1;
pjob->ji_qs.ji_svrflags &= ~JOB_SVFLG_StagedIn;
svr_evaljobstate(*pjob, newstate, newsubst, 0);
svr_setjobstate(pjob, newstate, newsubst, FALSE);
}
break;
} /* END switch (rc) */
pjob->ji_qs.ji_svrflags = (pjob->ji_qs.ji_svrflags &
~(JOB_SVFLG_CHECKPOINT_FILE |JOB_SVFLG_CHECKPOINT_MIGRATEABLE |
JOB_SVFLG_CHECKPOINT_COPIED)) | JOB_SVFLG_HASRUN;
sprintf(log_buf, msg_manager, msg_jobrerun, preq->rq_user, preq->rq_host);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
reply_ack(preq);
/* note in accounting file */
account_record(PBS_ACCT_RERUN, pjob, NULL);
return rc;
} /* END req_rerunjob() */
void purge_completed_jobs(
struct batch_request *preq) /* I */
{
job *pjob;
char *time_str;
time_t purge_time = 0;
int iter;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* get the time to purge the jobs that completed before */
time_str = preq->rq_extend;
time_str += strlen(PURGECOMP);
purge_time = strtol(time_str,NULL,10);
/*
* Clean unreported capability is only for operators and managers.
* Check if request is authorized
*/
if ((preq->rq_perm & (ATR_DFLAG_OPRD|ATR_DFLAG_OPWR|
ATR_DFLAG_MGRD|ATR_DFLAG_MGWR)) == 0)
{
req_reject(PBSE_PERM,0,preq,NULL,
"must have operator or manager privilege to use -c parameter");
return;
}
reply_ack(preq);
if (LOGLEVEL >= 4)
{
sprintf(log_buf,"Received purge completed jobs command, purge time is %ld (%s)",
(long)purge_time, preq->rq_extend);
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_REQUEST, __func__, log_buf);
}
iter = -1;
while ((pjob = next_job(&alljobs,&iter)) != NULL)
{
if ((pjob->ji_qs.ji_substate == JOB_SUBSTATE_COMPLETE) &&
(pjob->ji_wattr[JOB_ATR_comp_time].at_val.at_long <= purge_time) &&
((pjob->ji_wattr[JOB_ATR_reported].at_flags & ATR_VFLAG_SET) != 0) &&
(pjob->ji_wattr[JOB_ATR_reported].at_val.at_long == 0))
{
if (LOGLEVEL >= 4)
{
sprintf(log_buf,"Reported job is COMPLETED (%ld), setting reported to TRUE",
pjob->ji_wattr[JOB_ATR_comp_time].at_val.at_long);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
}
pjob->ji_wattr[JOB_ATR_reported].at_val.at_long = 1;
pjob->ji_wattr[JOB_ATR_reported].at_flags = ATR_VFLAG_SET | ATR_VFLAG_MODIFY;
job_save(pjob, SAVEJOB_FULL, 0);
}
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
}
return;
} /* END purge_completed_jobs() */
void delay_and_send_sig_kill(
batch_request *preq_sig)
{
int delay = 0;
job *pjob;
pbs_queue *pque;
batch_request *preq_clt = NULL; /* original client request */
int rc;
time_t time_now = time(NULL);
char log_buf[LOCAL_LOG_BUF_SIZE];
if (preq_sig == NULL)
return;
rc = preq_sig->rq_reply.brp_code;
if (preq_sig->rq_extend != NULL)
{
preq_clt = get_remove_batch_request(preq_sig->rq_extend);
}
/* the client request has been handled another way, nothing left to do */
if (preq_clt == NULL)
return;
if ((pjob = chk_job_request(preq_clt->rq_ind.rq_rerun, preq_clt)) == NULL)
{
/* job has gone away, chk_job_request() calls req_reject() on failure */
return;
}
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
if (rc)
{
/* mom rejected request */
if (rc == PBSE_UNKJOBID)
{
/* MOM claims no knowledge, so just purge it */
log_event(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
"MOM rejected signal during rerun");
/* removed the resources assigned to job */
free_nodes(pjob);
set_resc_assigned(pjob, DECR);
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
svr_job_purge(pjob);
reply_ack(preq_clt);
}
else
{
pjob_mutex.unlock();
req_reject(rc, 0, preq_clt, NULL, NULL);
}
return;
}
// Apply the user delay first so it takes precedence.
if (pjob->ji_wattr[JOB_ATR_user_kill_delay].at_flags & ATR_VFLAG_SET)
delay = pjob->ji_wattr[JOB_ATR_user_kill_delay].at_val.at_long;
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
mutex_mgr server_mutex = mutex_mgr(server.sv_attr_mutex, false);
if (delay == 0)
{
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
0);
}
}
else
{
/* why is the pque null. Something went wrong */
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "jobid %s returned a null queue", pjob->ji_qs.ji_jobid);
req_reject(PBSE_UNKQUE, 0, preq_clt, NULL, log_buf);
return;
}
pjob_mutex.unlock();
reply_ack(preq_clt);
set_task(WORK_Timed, delay + time_now, send_sig_kill, strdup(pjob->ji_qs.ji_jobid), FALSE);
} // END delay_and_send_sig_kill()
int req_rerunjob(
batch_request *preq)
{
int rc = PBSE_NONE;
job *pjob;
int MgrRequired = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* check if requestor is admin, job owner, etc */
if (!strcasecmp(preq->rq_ind.rq_rerun, "all"))
{
return(handle_requeue_all(preq));
}
if ((pjob = chk_job_request(preq->rq_ind.rq_rerun, preq)) == 0)
{
/* FAILURE */
/* chk_job_request calls req_reject() */
rc = PBSE_SYSTEM;
return rc; /* This needs to fixed to return an accurate error */
}
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
/* the job must be running or completed */
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
if (pjob->ji_wattr[JOB_ATR_checkpoint_name].at_flags & ATR_VFLAG_SET)
{
/* allow end-users to rerun checkpointed jobs */
MgrRequired = FALSE;
}
}
else if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* job is running */
/* NO-OP */
}
else if (pjob->ji_qs.ji_state == JOB_STATE_QUEUED)
{
// If we are already queued, then there is nothing to do.
rc = PBSE_NONE;
reply_ack(preq);
return(rc);
}
else
{
/* FAILURE - job is in bad state */
rc = PBSE_BADSTATE;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s is in a bad state",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
if ((MgrRequired == TRUE) &&
((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0))
{
/* FAILURE */
rc = PBSE_PERM;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"additional permissions required (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)");
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
/* the job must be rerunnable */
if (pjob->ji_wattr[JOB_ATR_rerunable].at_val.at_long == 0)
{
/* NOTE: should force override this constraint? maybe (???) */
/* no, the user is saying that the job will break, and
IEEE Std 1003.1 specifically says rerun is to be rejected
if rerunable==FALSE -garrick */
rc = PBSE_NORERUN;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s not rerunnable",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* ask MOM to kill off the job if it is running */
int delay = 0;
pbs_queue *pque;
// Apply the user delay first so it takes precedence.
if (pjob->ji_wattr[JOB_ATR_user_kill_delay].at_flags & ATR_VFLAG_SET)
delay = pjob->ji_wattr[JOB_ATR_user_kill_delay].at_val.at_long;
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
mutex_mgr server_mutex = mutex_mgr(server.sv_attr_mutex, false);
if (delay == 0)
{
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
0);
}
}
else
{
/* why is the pque null. Something went wrong */
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "jobid %s returned a null queue", pjob->ji_qs.ji_jobid);
req_reject(PBSE_UNKQUE, 0, preq, NULL, log_buf);
return(PBSE_UNKQUE);
}
pjob->ji_qs.ji_substate = JOB_SUBSTATE_RERUN;
if (delay != 0)
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
get_batch_request_id(preq);
/* If a qrerun -f is given requeue the job regardless of the outcome of issue_signal*/
if ((preq->rq_extend) &&
(!strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE))))
{
std::string extend = RERUNFORCE;
batch_request *dup = new batch_request(*preq);
get_batch_request_id(dup);
rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(dup->rq_id.c_str()));
if (rc == PBSE_NORELYMOM)
{
dup->rq_reply.brp_code = PBSE_NORELYMOM;
pjob_mutex.unlock();
post_rerun(dup);
pjob = svr_find_job(preq->rq_ind.rq_signal.rq_jid, FALSE);
if (pjob == NULL)
{
delete dup;
return(PBSE_NONE);
}
pjob_mutex.set_lock_state(true);
rc = PBSE_NONE;
}
delete dup;
}
else
{
rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(preq->rq_id.c_str()));
if (rc != PBSE_NONE)
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
return(rc);
}
}
else
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
/* If a qrerun -f is given requeue the job regardless of the outcome of issue_signal*/
if (preq->rq_extend && !strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE)))
{
std::string extend = RERUNFORCE;
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, strdup(extend.c_str()));
if (rc == PBSE_NORELYMOM)
rc = PBSE_NONE;
}
else
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, NULL);
}
}
else
{
if (pjob->ji_wattr[JOB_ATR_hold].at_val.at_long == HOLD_n)
{
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_QUEUED, FALSE);
}
else
{
svr_setjobstate(pjob, JOB_STATE_HELD, JOB_SUBSTATE_HELD, FALSE);
}
/* reset some job attributes */
pjob->ji_wattr[JOB_ATR_comp_time].at_flags &= ~ATR_VFLAG_SET;
pjob->ji_wattr[JOB_ATR_reported].at_flags &= ~ATR_VFLAG_SET;
set_statechar(pjob);
rc = -1;
}
/* finalize_rerunjob will return with pjob->ji_mutex unlocked */
pjob_mutex.set_unlock_on_exit(false);
return finalize_rerunjob(preq,pjob,rc);
}
int req_movejob(
batch_request *req) /* I */
{
job *jobp;
char log_buf[LOCAL_LOG_BUF_SIZE];
int local_errno = 0;
jobp = chk_job_request(req->rq_ind.rq_move.rq_jid, req);
if (jobp == NULL)
{
return(PBSE_NONE);
}
mutex_mgr job_mutex(jobp->ji_mutex, true);
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "%s", jobp->ji_qs.ji_jobid);
LOG_EVENT(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
}
if ((jobp->ji_qs.ji_state != JOB_STATE_QUEUED) &&
(jobp->ji_qs.ji_state != JOB_STATE_HELD) &&
(jobp->ji_qs.ji_state != JOB_STATE_WAITING))
{
#ifndef NDEBUG
sprintf(log_buf, "%s %d %s", pbse_to_txt(PBSE_BADSTATE), jobp->ji_qs.ji_state, __func__);
log_event(PBSEVENT_DEBUG,PBS_EVENTCLASS_JOB,jobp->ji_qs.ji_jobid,log_buf);
#endif /* NDEBUG */
req_reject(PBSE_BADSTATE, 0, req, NULL, NULL);
return(PBSE_NONE);
}
/*
* svr_movejob() does the real work, handles both local and
* network moves
*/
/* We have found that sometimes the destination queue and the
parent queue are the same. If so we do not need to do
anything else */
if (strcmp(jobp->ji_qs.ji_queue, req->rq_ind.rq_move.rq_destin) == 0)
{
sprintf(log_buf, "Job %s already in queue %s", jobp->ji_qs.ji_jobid, jobp->ji_qs.ji_queue);
if (LOGLEVEL >= 7)
{
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
}
req_reject(PBSE_JOB_ALREADY_IN_QUEUE, 0, req, NULL, log_buf);
return(PBSE_NONE);
}
switch (svr_movejob(jobp, req->rq_ind.rq_move.rq_destin, &local_errno, req))
{
case 0:
/* success */
snprintf(log_buf, sizeof(log_buf), "%s", msg_movejob);
snprintf(log_buf + strlen(log_buf), sizeof(log_buf) - strlen(log_buf), msg_manager,
req->rq_ind.rq_move.rq_destin, req->rq_user, req->rq_host);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,jobp->ji_qs.ji_jobid,log_buf);
reply_ack(req);
break;
case - 1:
case 1:
/* fail */
/* NOTE: can pass detailed response to requestor (NYI) */
req_reject(local_errno, 0, req, NULL, NULL);
break;
case 2:
/* deferred, will be handled by */
/* post_movejob() when the child completes */
/* NO-OP */
break;
} /* END switch (svr_movejob(jobp,req->rq_ind.rq_move.rq_destin,req)) */
return(PBSE_NONE);
} /* END req_movejob() */
int req_orderjob(
struct batch_request *vp) /* I */
{
job *pjob;
job *pjob1;
job *pjob2;
int rank;
int rc = 0;
char tmpqn[PBS_MAXQUEUENAME+1];
struct batch_request *req = (struct batch_request *)vp;
char log_buf[LOCAL_LOG_BUF_SIZE];
pbs_queue *pque1;
pbs_queue *pque2;
if ((pjob1 = chk_job_request(req->rq_ind.rq_move.rq_jid, req)) == NULL)
{
return(PBSE_NONE);
}
mutex_mgr job1_mutex(pjob1->ji_mutex, true);
if ((pjob2 = chk_job_request(req->rq_ind.rq_move.rq_destin, req)) == NULL)
{
return(PBSE_NONE);
}
mutex_mgr job2_mutex(pjob2->ji_mutex, true);
if (((pjob = pjob1)->ji_qs.ji_state == JOB_STATE_RUNNING) ||
((pjob = pjob2)->ji_qs.ji_state == JOB_STATE_RUNNING))
{
#ifndef NDEBUG
sprintf(log_buf, "%s %d",
pbse_to_txt(PBSE_BADSTATE),
pjob->ji_qs.ji_state);
strcat(log_buf, __func__);
log_event(
PBSEVENT_DEBUG,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buf);
#endif /* NDEBUG */
req_reject(PBSE_BADSTATE, 0, req, NULL, NULL);
return(PBSE_NONE);
}
else if ((pjob1->ji_qhdr == NULL) || (pjob2->ji_qhdr == NULL))
{
req_reject(PBSE_BADSTATE, 0, req, NULL, "One of the jobs does not have a queue");
return(PBSE_NONE);
}
else if (pjob1->ji_qhdr != pjob2->ji_qhdr)
{
/* jobs are in different queues */
int ok = FALSE;
if ((pque2 = get_jobs_queue(&pjob2)) == NULL)
{
rc = PBSE_BADSTATE;
job2_mutex.set_lock_on_exit(false);
}
else
{
mutex_mgr pque2_mutex = mutex_mgr(pque2->qu_mutex, true);
if ((rc = svr_chkque(pjob1, pque2, get_variable(pjob1, pbs_o_host), MOVE_TYPE_Order, NULL)) == PBSE_NONE)
{
pque2_mutex.unlock();
if ((pque1 = get_jobs_queue(&pjob1)) == NULL)
{
rc = PBSE_BADSTATE;
job1_mutex.set_lock_on_exit(false);
}
else if (pjob1 != NULL)
{
mutex_mgr pque1_mutex = mutex_mgr(pque1->qu_mutex, true);
if ((rc = svr_chkque(pjob2, pque1, get_variable(pjob2, pbs_o_host), MOVE_TYPE_Order, NULL)) == PBSE_NONE)
{
ok = TRUE;
}
}
}
}
if (ok == FALSE)
{
req_reject(rc, 0, req, NULL, NULL);
return(PBSE_NONE);
}
}
/* now swap the order of the two jobs in the queue lists */
rank = pjob1->ji_wattr[JOB_ATR_qrank].at_val.at_long;
pjob1->ji_wattr[JOB_ATR_qrank].at_val.at_long =
pjob2->ji_wattr[JOB_ATR_qrank].at_val.at_long;
pjob2->ji_wattr[JOB_ATR_qrank].at_val.at_long = rank;
if (pjob1->ji_qhdr != pjob2->ji_qhdr)
{
strcpy(tmpqn, pjob1->ji_qs.ji_queue);
strcpy(pjob1->ji_qs.ji_queue, pjob2->ji_qs.ji_queue);
strcpy(pjob2->ji_qs.ji_queue, tmpqn);
svr_dequejob(pjob1, FALSE);
svr_dequejob(pjob2, FALSE);
if (svr_enquejob(pjob1, FALSE, -1) == PBSE_JOB_RECYCLED)
{
pjob1 = NULL;
job1_mutex.set_lock_on_exit(false);
}
if (svr_enquejob(pjob2, FALSE, -1) == PBSE_JOB_RECYCLED)
{
pjob2 = NULL;
job2_mutex.set_lock_on_exit(false);
}
}
else
{
if ((pque1 = get_jobs_queue(&pjob1)) != NULL)
{
mutex_mgr pque1_mutex = mutex_mgr(pque1->qu_mutex, true);
swap_jobs(pque1->qu_jobs,pjob1,pjob2);
swap_jobs(NULL,pjob1,pjob2);
}
}
/* need to update disk copy of both jobs to save new order */
if (pjob1 != NULL)
{
job_save(pjob1, SAVEJOB_FULL, 0);
}
if (pjob2 != NULL)
{
job_save(pjob2, SAVEJOB_FULL, 0);
}
/* SUCCESS */
reply_ack(req);
return(PBSE_NONE);
} /* END req_orderjob() */
void req_deletearray(struct batch_request *preq)
{
job_array *pa;
char *range;
struct work_task *ptask;
int num_skipped;
char owner[PBS_MAXUSER + 1];
pa = get_array(preq->rq_ind.rq_delete.rq_objname);
if (pa == NULL)
{
reply_ack(preq);
return;
}
/* check authorization */
get_jobowner(pa->ai_qs.owner, owner);
if (svr_authorize_req(preq, owner, pa->ai_qs.submit_host) == -1)
{
sprintf(log_buffer, msg_permlog,
preq->rq_type,
"Array",
preq->rq_ind.rq_delete.rq_objname,
preq->rq_user,
preq->rq_host);
log_event(
PBSEVENT_SECURITY,
PBS_EVENTCLASS_JOB,
preq->rq_ind.rq_delete.rq_objname,
log_buffer);
req_reject(PBSE_PERM, 0, preq, NULL, "operation not permitted");
return;
}
/* get the range of jobs to iterate over */
range = preq->rq_extend;
if ((range != NULL) &&
(strstr(range,ARRAY_RANGE) != NULL))
{
/* parse the array range */
num_skipped = delete_array_range(pa,range);
if (num_skipped < 0)
{
/* ERROR */
req_reject(PBSE_IVALREQ,0,preq,NULL,"Error in specified array range");
return;
}
}
else
{
num_skipped = delete_whole_array(pa);
}
/* check if the array is gone */
if ((pa = get_array(preq->rq_ind.rq_delete.rq_objname)) != NULL)
{
/* some jobs were not deleted. They must have been running or had
JOB_SUBSTATE_TRANSIT */
if (num_skipped != 0)
{
ptask = set_task(WORK_Timed, time_now + 2, array_delete_wt, preq);
if(ptask)
{
return;
}
}
}
/* now that the whole array is deleted, we should mail the user if necessary */
reply_ack(preq);
return;
}
static void post_movejob(
struct work_task *pwt)
{
char *id = "post_movejob";
struct batch_request *req;
int newstate;
int newsub;
int stat;
int r;
job *jobp;
req = (struct batch_request *)pwt->wt_parm2;
stat = pwt->wt_aux;
pbs_errno = PBSE_NONE;
if (req->rq_type != PBS_BATCH_MoveJob)
{
sprintf(log_buffer, "bad request type %d\n",
req->rq_type);
log_err(-1, id, log_buffer);
return;
}
jobp = find_job(req->rq_ind.rq_move.rq_jid);
if ((jobp == NULL) || (jobp != (job *)pwt->wt_parm1))
{
sprintf(log_buffer, "job %s not found\n",
req->rq_ind.rq_move.rq_jid);
log_err(-1, id, log_buffer);
}
if (WIFEXITED(stat))
{
r = WEXITSTATUS(stat);
if (r == 0)
{
/* purge server's job structure */
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(jobp);
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_COPIED)
remove_checkpoint(jobp);
strcpy(log_buffer, msg_movejob);
sprintf(log_buffer + strlen(log_buffer), msg_manager,
req->rq_ind.rq_move.rq_destin,
req->rq_user,
req->rq_host);
job_purge(jobp);
}
else
{
r = PBSE_ROUTEREJ;
}
}
else
{
r = PBSE_SYSTEM;
sprintf(log_buffer, msg_badexit, stat);
strcat(log_buffer, id);
log_event(
PBSEVENT_SYSTEM,
PBS_EVENTCLASS_JOB,
jobp->ji_qs.ji_jobid,
log_buffer);
}
if (r)
{
if (jobp != NULL)
{
/* force re-eval of job state out of Transit */
svr_evaljobstate(jobp, &newstate, &newsub, 1);
svr_setjobstate(jobp, newstate, newsub);
}
req_reject(r, 0, req, NULL, NULL);
}
else
{
reply_ack(req);
}
return;
} /* END post_movejob() */
void
req_relnodesjob(struct batch_request *preq)
{
int jt; /* job type */
job *pjob;
int rc;
char *jid;
int i, offset;
char *nodeslist = NULL;
char msg[LOG_BUF_SIZE];
if (preq == NULL)
return;
jid = preq->rq_ind.rq_relnodes.rq_jid;
if (jid == NULL)
return;
/*
** Returns job pointer for singleton job or "parent" of
** an array job.
*/
pjob = chk_job_request(jid, preq, &jt);
if (pjob == NULL) {
return;
}
if (jt == IS_ARRAY_NO) { /* a regular job is okay */
/* the job must be running */
if ((pjob->ji_qs.ji_state != JOB_STATE_RUNNING) ||
(pjob->ji_qs.ji_substate !=
JOB_SUBSTATE_RUNNING)) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
}
else if (jt == IS_ARRAY_Single) { /* a single subjob is okay */
offset = subjob_index_to_offset(pjob,
get_index_from_jid(jid));
if (offset == -1) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
i = get_subjob_state(pjob, offset);
if (i == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
}
if (i != JOB_STATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
if ((pjob = pjob->ji_ajtrk->tkm_tbl[offset].trk_psubjob) == NULL) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
if (pjob->ji_qs.ji_substate != JOB_SUBSTATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
} else {
reply_text(preq, PBSE_NOSUP,
"not supported for Array Jobs or multiple sub-jobs");
return;
}
nodeslist = preq->rq_ind.rq_relnodes.rq_node_list;
if ((nodeslist != NULL) && (nodeslist[0] == '\0')) {
nodeslist = NULL;
}
rc = free_sister_vnodes(pjob, nodeslist, msg, LOG_BUF_SIZE, preq);
if (rc != 0) {
reply_text(preq, PBSE_SYSTEM, msg);
}
}
int req_stat_node(
struct batch_request *preq)
{
char *name;
int rc = PBSE_NONE;
int type = 0;
int bad = 0;
struct pbsnode *pnode = NULL;
struct batch_reply *preply;
struct prop props;
svrattrl *pal;
/*
* first, check that the server indeed has a list of nodes
* and if it does, validate the name of the requested object--
* either name is that of a specific node, or name[0] is null/@
* meaning request is for all nodes in the server's jurisdiction
*/
if (LOGLEVEL >= 6)
{
log_record( PBSEVENT_SCHED, PBS_EVENTCLASS_REQUEST, __func__, "entered");
}
if (svr_totnodes <= 0)
{
rc = PBSE_NONODES;
req_reject(rc, 0, preq, NULL, "node list is empty - check 'server_priv/nodes' file");
return rc;
}
name = preq->rq_ind.rq_status.rq_id;
if ((*name == '\0') || (*name == '@'))
{
type = 1;
}
else if ((*name == ':') && (*(name + 1) != '\0'))
{
if (!strcmp(name + 1, "ALL"))
{
type = 1; /* psuedo-group for all nodes */
}
else
{
type = 2;
props.name = name + 1;
props.mark = 1;
props.next = NULL;
}
}
preply = &preq->rq_reply;
preply->brp_choice = BATCH_REPLY_CHOICE_Status;
CLEAR_HEAD(preply->brp_un.brp_status);
if (type == 0)
{
/* get status of the named node */
pnode = find_nodebyname(name);
if (pnode == NULL)
{
rc = PBSE_UNKNODE;
req_reject(rc, 0, preq, NULL, "cannot locate specified node");
return(rc);
}
/* get the status on all of the numa nodes */
if (pnode->nd_is_alps_reporter == TRUE)
rc = get_alps_statuses(pnode, preq, &bad, &preply->brp_un.brp_status);
else
rc = get_numa_statuses(pnode, preq, &bad, &preply->brp_un.brp_status);
unlock_node(pnode, __func__, "type == 0", LOGLEVEL);
}
else
{
/* get status of all or several nodes */
all_nodes_iterator *iter = NULL;
while ((pnode = next_host(&allnodes,&iter,NULL)) != NULL)
{
if ((type == 2) &&
(!hasprop(pnode, &props)))
{
unlock_node(pnode, __func__, "type != 0, next_host", LOGLEVEL);
continue;
}
/* get the status on all of the numa nodes */
if (pnode->nd_is_alps_reporter == TRUE)
rc = get_alps_statuses(pnode, preq, &bad, &preply->brp_un.brp_status);
else
rc = get_numa_statuses(pnode, preq, &bad, &preply->brp_un.brp_status);
if (rc != PBSE_NONE)
{
unlock_node(pnode, __func__, "type != 0, rc != 0, get_numa_statuses", LOGLEVEL);
break;
}
unlock_node(pnode, __func__, "type != 0, rc == 0, get_numa_statuses", LOGLEVEL);
}
if (iter != NULL)
delete iter;
}
if (rc == PBSE_NONE)
{
/* SUCCESS */
reply_send_svr(preq);
}
else
{
if (rc != PBSE_UNKNODEATR)
{
req_reject(rc, 0, preq, NULL, NULL);
}
else
{
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
reply_badattr(rc, bad, pal, preq);
}
}
return(rc);
} /* END req_stat_node() */
int req_deletejob(
struct batch_request *preq) /* I */
{
char *Msg = NULL;
struct batch_request *preq_tmp = NULL;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* check if we are getting a purgecomplete from scheduler */
if (preq->rq_extend != NULL)
{
if (!strncmp(preq->rq_extend,PURGECOMP,strlen(PURGECOMP)))
{
/* purge_completed_jobs will respond with either an ack or reject */
purge_completed_jobs(preq);
return(PBSE_NONE);
}
else if (strncmp(preq->rq_extend, deldelaystr, strlen(deldelaystr)) &&
strncmp(preq->rq_extend, delasyncstr, strlen(delasyncstr)) &&
strncmp(preq->rq_extend, delpurgestr, strlen(delpurgestr)))
{
/* have text message in request extension, add it */
Msg = preq->rq_extend;
/* Message capability is only for operators and managers.
* Check if request is authorized */
if ((preq->rq_perm & (ATR_DFLAG_OPRD | ATR_DFLAG_OPWR |
ATR_DFLAG_MGRD | ATR_DFLAG_MGWR)) == 0)
{
req_reject(PBSE_PERM, 0, preq, NULL,
"must have operator or manager privilege to use -m parameter");
return(PBSE_NONE);
}
}
/* check if we are getting a asynchronous delete */
else if (!strncmp(preq->rq_extend,delasyncstr,strlen(delasyncstr)))
{
/*
* Respond with an ack now instead of after MOM processing
* Create a new batch request and fill it in. It will be freed by reply_ack
*/
snprintf(log_buf,sizeof(log_buf), "Deleting job asynchronously");
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,preq->rq_ind.rq_delete.rq_objname,log_buf);
preq_tmp = duplicate_request(preq);
}
}
if (strcasecmp(preq->rq_ind.rq_delete.rq_objname,"all") == 0)
{
handle_delete_all(preq, preq_tmp, Msg);
}
else
{
handle_single_delete(preq, preq_tmp, Msg);
}
return(PBSE_NONE);
} /* END req_deletejob() */
int req_stat_job(
struct batch_request *preq) /* ptr to the decoded request */
{
struct stat_cntl *cntl; /* see svrfunc.h */
char *name;
job *pjob = NULL;
pbs_queue *pque = NULL;
int rc = PBSE_NONE;
char log_buf[LOCAL_LOG_BUF_SIZE];
enum TJobStatTypeEnum type = tjstNONE;
/*
* first, validate the name of the requested object, either
* a job, a queue, or the whole server.
*/
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "note");
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
}
/* FORMAT: name = { <JOBID> | <QUEUEID> | '' } */
name = preq->rq_ind.rq_status.rq_id;
if (preq->rq_extend != NULL)
{
/* evaluate pbs_job_stat() 'extension' field */
if (!strncasecmp(preq->rq_extend, "truncated", strlen("truncated")))
{
/* truncate response by 'max_report' */
type = tjstTruncatedServer;
}
else if (!strncasecmp(preq->rq_extend, "summarize_arrays", strlen("summarize_arrays")))
{
type = tjstSummarizeArraysServer;
}
} /* END if (preq->rq_extend != NULL) */
if (isdigit((int)*name))
{
/* status a single job */
if (is_array(name))
{
if (type != tjstSummarizeArraysServer)
{
type = tjstArray;
}
}
else
{
type = tjstJob;
if ((pjob = svr_find_job(name, FALSE)) == NULL)
{
rc = PBSE_UNKJOBID;
}
else
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
}
}
else if (isalpha(name[0]))
{
if (type == tjstNONE)
type = tjstQueue;
else if (type == tjstSummarizeArraysServer)
type = tjstSummarizeArraysQueue;
else
type = tjstTruncatedQueue;
/* if found, this mutex is released later */
if ((pque = find_queuebyname(name)) == NULL)
{
rc = PBSE_UNKQUE;
}
}
else if ((*name == '\0') || (*name == '@'))
{
/* status all jobs at server */
if (type == tjstNONE)
type = tjstServer;
}
else
{
rc = PBSE_IVALREQ;
}
if (rc != 0)
{
/* is invalid - an error */
req_reject(rc, 0, preq, NULL, NULL);
return(rc);
}
preq->rq_reply.brp_choice = BATCH_REPLY_CHOICE_Status;
CLEAR_HEAD(preq->rq_reply.brp_un.brp_status);
cntl = (struct stat_cntl *)calloc(1, sizeof(struct stat_cntl));
if (cntl == NULL)
{
if (pque != NULL)
unlock_queue(pque, "req_stat_job", (char *)"no memory cntl", LOGLEVEL);
req_reject(PBSE_SYSTEM, 0, preq, NULL, NULL);
return(PBSE_SYSTEM);
}
if ((type == tjstTruncatedQueue) ||
(type == tjstTruncatedServer))
{
if (pque != NULL)
{
unlock_queue(pque, __func__, "", LOGLEVEL);
pque = NULL;
}
}
cntl->sc_type = (int)type;
cntl->sc_conn = -1;
cntl->sc_pque = pque;
cntl->sc_origrq = preq;
cntl->sc_post = req_stat_job_step2;
cntl->sc_jobid[0] = '\0'; /* cause "start from beginning" */
req_stat_job_step2(cntl); /* go to step 2, see if running is current */
if (pque != NULL)
unlock_queue(pque, "req_stat_job", (char *)"success", LOGLEVEL);
free(cntl);
return(PBSE_NONE);
} /* END req_stat_job() */
int req_stat_svr(
struct batch_request *preq) /* ptr to the decoded request */
{
svrattrl *pal;
struct batch_reply *preply;
struct brp_status *pstat;
int bad = 0;
char nc_buf[128];
int numjobs;
int netrates[3];
memset(netrates, 0, sizeof(netrates));
/* update count and state counts from sv_numjobs and sv_jobstates */
lock_sv_qs_mutex(server.sv_qs_mutex, __func__);
numjobs = server.sv_qs.sv_numjobs;
unlock_sv_qs_mutex(server.sv_qs_mutex, __func__);
pthread_mutex_lock(server.sv_attr_mutex);
server.sv_attr[SRV_ATR_TotalJobs].at_val.at_long = numjobs;
server.sv_attr[SRV_ATR_TotalJobs].at_flags |= ATR_VFLAG_SET;
pthread_mutex_lock(server.sv_jobstates_mutex);
update_state_ct(
&server.sv_attr[SRV_ATR_JobsByState],
server.sv_jobstates,
server.sv_jobstbuf);
pthread_mutex_unlock(server.sv_jobstates_mutex);
netcounter_get(netrates);
snprintf(nc_buf, 127, "%d %d %d", netrates[0], netrates[1], netrates[2]);
if (server.sv_attr[SRV_ATR_NetCounter].at_val.at_str != NULL)
free(server.sv_attr[SRV_ATR_NetCounter].at_val.at_str);
server.sv_attr[SRV_ATR_NetCounter].at_val.at_str = strdup(nc_buf);
if (server.sv_attr[SRV_ATR_NetCounter].at_val.at_str != NULL)
server.sv_attr[SRV_ATR_NetCounter].at_flags |= ATR_VFLAG_SET;
pthread_mutex_unlock(server.sv_attr_mutex);
/* allocate a reply structure and a status sub-structure */
preply = &preq->rq_reply;
preply->brp_choice = BATCH_REPLY_CHOICE_Status;
CLEAR_HEAD(preply->brp_un.brp_status);
pstat = (struct brp_status *)calloc(1, sizeof(struct brp_status));
if (pstat == NULL)
{
reply_free(preply);
req_reject(PBSE_SYSTEM, 0, preq, NULL, NULL);
pthread_mutex_unlock(server.sv_attr_mutex);
return(PBSE_SYSTEM);
}
CLEAR_LINK(pstat->brp_stlink);
strcpy(pstat->brp_objname, server_name);
pstat->brp_objtype = MGR_OBJ_SERVER;
CLEAR_HEAD(pstat->brp_attr);
append_link(&preply->brp_un.brp_status, &pstat->brp_stlink, pstat);
/* add attributes to the status reply */
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
if (status_attrib(
pal,
svr_attr_def,
server.sv_attr,
SRV_ATR_LAST,
preq->rq_perm,
&pstat->brp_attr,
&bad,
1)) /* IsOwner == TRUE */
{
reply_badattr(PBSE_NOATTR, bad, pal, preq);
}
else
{
reply_send_svr(preq);
}
return(PBSE_NONE);
} /* END req_stat_svr() */
static void post_delete_mom1(
struct work_task *pwt)
{
int delay = 0;
int dellen = strlen(deldelaystr);
job *pjob;
pbs_queue *pque;
char *preq_clt_id;
struct batch_request *preq_sig; /* signal request to MOM */
struct batch_request *preq_clt = NULL; /* original client request */
int rc;
time_t time_now = time(NULL);
preq_sig = get_remove_batch_request((char *)pwt->wt_parm1);
free(pwt->wt_mutex);
free(pwt);
if (preq_sig == NULL)
return;
rc = preq_sig->rq_reply.brp_code;
preq_clt_id = preq_sig->rq_extra;
free_br(preq_sig);
if (preq_clt_id != NULL)
{
preq_clt = get_remove_batch_request(preq_clt_id);
free(preq_clt_id);
}
/* the client request has been handled another way, nothing left to do */
if (preq_clt == NULL)
return;
pjob = svr_find_job(preq_clt->rq_ind.rq_delete.rq_objname, FALSE);
if (pjob == NULL)
{
/* job has gone away */
req_reject(PBSE_UNKJOBID, 0, preq_clt, NULL, NULL);
return;
}
if (rc)
{
/* mom rejected request */
if (rc == PBSE_UNKJOBID)
{
/* MOM claims no knowledge, so just purge it */
log_event(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
"MOM rejected signal during delete");
/* removed the resources assigned to job */
free_nodes(pjob);
set_resc_assigned(pjob, DECR);
svr_job_purge(pjob);
reply_ack(preq_clt);
}
else
{
req_reject(rc, 0, preq_clt, NULL, NULL);
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
}
return;
}
if (preq_clt->rq_extend)
{
if (strncmp(preq_clt->rq_extend, deldelaystr, dellen) == 0)
{
delay = atoi(preq_clt->rq_extend + dellen);
}
}
reply_ack(preq_clt); /* dont need it, reply now */
/*
* if no delay specified in original request, see if kill_delay
* queue attribute is set.
*/
if (delay == 0)
{
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
pthread_mutex_lock(server.sv_attr_mutex);
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
2);
pthread_mutex_unlock(server.sv_attr_mutex);
unlock_queue(pque, __func__, NULL, LOGLEVEL);
}
else if (pjob != NULL)
return;
}
set_task(WORK_Timed, delay + time_now, post_delete_mom2, strdup(pjob->ji_qs.ji_jobid), FALSE);
/*
* Since the first signal has succeeded, let's reschedule the
* nanny to be 1 minute after the second phase.
*/
apply_job_delete_nanny(pjob, time_now + delay + 60);
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
} /* END post_delete_mom1() */
void
req_py_spawn(struct batch_request *preq)
{
int jt; /* job type */
job *pjob;
int rc;
char *jid = preq->rq_ind.rq_py_spawn.rq_jid;
int i, offset;
/*
** Returns job pointer for singleton job or "parent" of
** an array job.
*/
pjob = chk_job_request(jid, preq, &jt);
if (pjob == NULL)
return;
/* see if requestor is the job owner */
if (svr_chk_owner(preq, pjob) != 0) {
req_reject(PBSE_PERM, 0, preq);
return;
}
if (jt == IS_ARRAY_NO) { /* a regular job is okay */
/* the job must be running */
if ((pjob->ji_qs.ji_state != JOB_STATE_RUNNING) ||
(pjob->ji_qs.ji_substate !=
JOB_SUBSTATE_RUNNING)) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
}
else if (jt == IS_ARRAY_Single) { /* a single subjob is okay */
offset = subjob_index_to_offset(pjob,
get_index_from_jid(jid));
if (offset == -1) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
i = get_subjob_state(pjob, offset);
if (i == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
}
if (i != JOB_STATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
if ((pjob = pjob->ji_ajtrk->tkm_tbl[offset].trk_psubjob) == NULL) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
if (pjob->ji_qs.ji_substate != JOB_SUBSTATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
} else {
reply_text(preq, PBSE_NOSUP,
"not supported for Array Jobs or multiple sub-jobs");
return;
}
/*
** Pass the request on to MOM. If this works, the function
** post_py_spawn_req will be called to handle the reply.
** If it fails, send the reply now.
*/
rc = relay_to_mom(pjob, preq, post_py_spawn_req);
if (rc)
req_reject(rc, 0, preq); /* unable to get to MOM */
}